[ https://issues.apache.org/jira/browse/CASSANDRA-13114?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15851430#comment-15851430 ]
Stefan Podkowinski commented on CASSANDRA-13114: ------------------------------------------------ Any exceptions caught be Netty will be logged by {{UnexpectedChannelExceptionHandler}}. Before the changes to {{Message}}, this would only happen when the channel would still be open. In Netty 4.0.23 this seems to be the case, in 4.0.44 the channel is now closed when {{exceptionCaught}} is called. What the patch does is simply to log the exception in both cases, while leaving the error reply sending case as it was (keep sending reply only in case channel is still open). I've noticed the changed behaviour through a dtest, [see results|http://cassci.datastax.com/view/Dev/view/spodkowinski/job/spodkowinski-CASSANDRA-13114-3.11-dtest/1/testReport/junit/native_transport_ssl_test/NativeTransportSSL/connect_to_ssl_test/]. As for the mentioned performance regression tests, I've only have a single physical host available for cstar_perf, so I can't really stress test multi-node setups. A single node tests with and without ssl didn't show any significant changes, but the results are little meaningful. Besides I also can't find any way to test the actual memory leak issue at hand by simulating lots of connecting and deconnecting clients. > 3.0.x: update netty > ------------------- > > Key: CASSANDRA-13114 > URL: https://issues.apache.org/jira/browse/CASSANDRA-13114 > Project: Cassandra > Issue Type: Bug > Reporter: Tom van der Woerdt > Assignee: Stefan Podkowinski > Attachments: 13114_netty-4.0.44_2.x-3.0.patch, > 13114_netty-4.0.44_3.11.patch > > > https://issues.apache.org/jira/browse/CASSANDRA-12032 updated netty for > Cassandra 3.8, but this wasn't backported. Netty 4.0.23, which ships with > Cassandra 3.0.x, has some serious bugs around memory handling for SSL > connections. > It would be nice if both were updated to 4.0.42, a version released this year. > 4.0.23 makes it impossible for me to run SSL, because nodes run out of memory > every ~30 minutes. This was fixed in 4.0.27. -- This message was sent by Atlassian JIRA (v6.3.15#6346)